Dynamic Object Detection and Segmentation Based on Mask R-CNN

In real situations, problems such as false detection or omission can occur in dynamic scene object detection, so this document aims to solve this problem. This paper proposes a new dynamic object detection and segmentation method based on improvements to the Mask R-CNN algorithm. First, the Mask-R-CNN algorithm is improved to introduce temporal information to improve model robustness through temporal context fusion. The network structure is then optimized, and multi-layered pyramid feature fusion and hierarchical prediction mechanisms are used to improve model accuracy and speed. Experimental results show that the proposed method performs better in dynamic object detection and segmentation tasks.